Autism spectrum disorders (ASD) are characterized by deficits in social interaction and communication, and unusually repetitive and restricted behaviors and interests (American Psychiatric Association, 2013). Reliable instruments are needed to quantify autistic traits and screen for high-risk children, thereby improving early detection of autism and promoting better emotional and behavioral outcomes (de Leeuw et al., 2020; Saito et al., 2017). However, the majority of screening tools for autistic traits have been developed in Western high-income countries, typically in Western Europe and North America (Durkin et al., 2015; West et al., 2016). Our knowledge of measurements of autistic traits may thus be culturally and contextually biased (Durkin et al., 2015). In this study, we compare a popular parent-report screening tool for ASD between children with ASD from Eastern (Chinese) and Western (The Netherlands) cultures.

If the recognition, interpretation and reporting of children’s autistic traits are not consistent across cultures, as suggested by a recent review (de Leeuw et al., 2020), this restricts the usability of autism screening tools developed in Western cultures in other countries (Norbury & Sparks, 2013). Even though there is a growing number of studies attempting to develop or adapt screening instruments for use in Eastern cultures, outcomes are not always optimal. For instance, translated screening instruments showed low sensitivity (0.48) in identifying children with ASD in Japan (Kamio et al., 2014), and had low to moderate internal consistencies in China (0.21–0.69) (Zhang et al., 2018). There is preliminary evidence suggesting subtle differences in autistic traits across Eastern and Western countries (de Leeuw et al., 2020; Freeth et al., 2014; Norbury & Sparks, 2013). For example, disinterest toward other children, which is an important indicator of autism in the original screening tools developed in the United States (Robins et al., 2001), is not reported by Japanese caregivers as an autism symptom of their children with ASD (Inada et al., 2011). As Japanese people generally are more introverted than westerners (Wakabayashi et al., 2006a, 2006b), Japanese caregivers may interpret a lack of interest in other children as modesty or shyness rather than an autistic trait. Therefore, assessment tools that have been developed and standardized in the West may not be reliable or valid if applied in other non-Western cultures (Henrich et al., 2010). In line with this, a lack of culturally adapted screening tools may further contribute to the discrepancies in prevalence estimates (Pang et al., 2018). China, with a population of almost 1.4 billion and an increasing prevalence of ASD, represents an enormously growing population of individuals with ASD. Yet, the prevalence rate of autism in Chinese children is reported to be 0.7% (Zhou et al., 2020), which is lower than the 1–2% prevalence rates among Western children (Baron-Cohen et al., 2009; Xu et al., 2018). Thus, further research is needed to evaluate whether autism screening tools are culturally appropriate and whether they are valid across different countries and cultures. So far, there have been no systematic comparisons of children’s autistic traits from Chinese and Western samples.

Besides the role of culture, socioeconomic status (SES) and demographic (sex and age) information may also impact on the recognition, interpretation and reporting of autistic traits. Parents from low SES families (e.g., lower maternal education) within the United States tend to over-report symptoms associated with ASD (Khowaja et al., 2014; Scarpa et al., 2013) and exhibit lower consistency in screening outcomes when compared to families with higher maternal education levels (Khowaja et al., 2014). This high false-positive rate might be due to the limited awareness of early childhood development and behaviors in families with low SES (Bishop-Fitzpatrick & Kind, 2017; Colbert et al., 2017; Ratto et al., 2016) or lack of internal consistency and accuracy in screening tools across low SES groups (Scarpa et al., 2013). Whether the effects of SES on parental report of children’s autistic traits is the same in different cultural contexts is still unknown. Within ASD, males outnumber females with a sex ratio of 4.3:1 (Maenner, 2020). Previous studies found that boys with ASD are prone to isolation and thus less social (Hiller et al., 2015) and have more stereotyped and repetitive behaviors than girls (Hiller et al., 2015; Sun et al., 2019; Szatmari et al., 2012). Moreover, although ASD is generally considered a ‘life-long’ condition, the severity of core symptoms seems to increase or decrease as children grow older (Fountain et al., 2012; Louwerse et al., 2015; Scheeren et al., 2019). Increased insight into the effects of SES, sex and age on autistic traits in children with ASD will promote screening and diagnostic accuracy.

Worldwide, the most frequently used questionnaire for autism is the Autism-spectrum Quotient (AQ; Baron-Cohen et al., 2001), which is a 50-item self-report or parent-report measure. The parent-report children’s version of the AQ (AQ-Child) aims to quantify autistic traits in children 4–11 years old across five theoretical domains (Social skills, Attention to detail, Attention switching, Communication and Imagination) (Auyeung et al., 2007; Baron-Cohen et al., 2001). The factor structure of the parent-report AQ-Child varies according to different studies (Auyeung et al., 2007; Gomez et al., 2018; Sun et al., 2019). Auyeung et al. (2007) found a 4-factor model (Mind-reading, Attention to detail, Social Skills and Imagination), with 47 items based on 1225 typically developing UK children. A Chinese study reported a 5-factor model (Socialness, Social communicative competence, Imagination, Patterns and Attention Switching), with 30 items in a sample of 1020 typically developing children (Sun et al., 2019). However, they reported a low alpha coefficient (0.588) for the Attention Switching domain, suggesting weak internal consistency of this subscale. Finally, Gomez et al. (2018) found a 4-factor model with 32 items reflecting Mind-reading, Social Skills, Attention to details and Imagination in 404 Australian children with a diagnosis of Attention Deficit/Hyperactivity Disorder (ADHD). These inconsistent factor structures raise the concern that the construct validity and latent structure of the AQ-Child may differ across samples, countries and cultures. These studies have all been conducted in samples that are typically developing or had additional diagnoses such as ADHD. Therefore, there is a need for evaluating the structure of this parent report measure within a sample of children with ASD.

A popular, shortened version, the Autism Spectrum Quotient-Short Form (AQ-Short, Hoekstra et al., 2011), includes 28 items targeting the core dimensions of autism and is suitable for quick assessment of autistic traits in large samples (Murray et al., 2013; Rentergem et al., 2019). Previous research has supported a two-factor structure, including a social behavior factor and a numbers and patterns factor, of the self-report AQ-Short in adults with ASD (Grove et al., 2016; Kuenssberg et al., 2014; Murray et al., 2013). While these results indicate that the AQ-Short is a valid and useful autism screening instrument in adults with ASD, the AQ-Short has not been tested as a parent report measure of autistic traits in children with ASD. As mentioned above, previous research evaluating the factor structure of the AQ-Child suggests a large degree of variability across samples and cultures. The reliability and validity of the parent-report AQ in samples of children with ASD needs more investigation. Moreover, underlying differences in the interpretation and reporting of autistic traits in different countries may affect the generalizability of Western screening tools in Eastern cultures. Clearly, there remains a need for structural validation studies to investigate whether the AQ-Short can be adapted as a parent report measure for use across cultures.

The current study’s objectives are threefold: (a) examine the factor structure of the parent-report version of the AQ-Short for children with ASD in China and the Netherlands; (b) compare autistic traits between children with ASD in China and the Netherlands; (c) explore the role of SES, children’s gender and age on parent-reported autistic traits in China and the Netherlands. As specified in our preregistered analysis plan (authors blinded), we expect that (1) the underlying factor structures of the AQ-Short for each population are similar; (2) more autistic traits will be reported for the Chinese compared to the Dutch participants; and (3) boys will be reported to score higher than girls on autistic traits in both countries.

Method

Study Sample

The current study included parent-reported data on children with ASD from China (n = 327, 85% boys, 3 to 17 years) and the Netherlands (n = 694, 78% boys, 6 to 16 years). The Chinese sample (n = 327) was recruited through the China Association of Persons with Psychiatric Disability and their Relatives (CAPPDR), the national autism organization in charge of providing services for ASD in China. Members of the CAPPDR service centers from all over China invited their patients and families to participate in our study. This data collection was conducted online in late 2018. Children diagnosed with ASD or receiving an ASD subsidy from the government were included in the final sample. Data of Dutch children with ASD (n = 694) were collected via the Netherlands Autism Register (NAR, https://www.nederlandsautismeregister.nl/english/), an online database that collects information from individuals with ASD and their families. The NAR data have been collected in annual waves since 2013 and cover various domains including general demographics, clinical diagnosis, co-occurring conditions, and autistic traits. All participants had a formal ASD diagnosis, the sample did not include control participants.

Measurements

Background information of the children and the parents was collected using structured questions on demographics (age, sex) and socioeconomic status (parents’ employment status and family income). Parents reported their employment status in three categories: employed, unemployed and other. Income data were available for 309 Chinese and 143 Dutch parents. Chinese parents reported their monthly income, whereas Dutch parents selected one of eight predefined yearly income levels. Since the two countries differ in absolute income, the categorization of income levels was based on country-specific percentiles (low: below 30th percentile; middle: 30–70th percentile; high: above 70th percentile) of the data from the National Bureau of Statistics of China (2019) and Statistic Netherlands (National CBS, 2019) separately. We categorized family income into low (China: 0–¥3590 a month, €0–€5532 a year; Netherlands: €0–€30,000), middle (China: ¥3590–¥9118 a month, €5532–€14,052 a year; Netherlands: €30,000–€70,000), and high income (China: above ¥9118, above €14,052 a year; Netherlands: > €70,000).

The Autism Spectrum Quotient-Short

The parent-report version of the Autism Spectrum Quotient Short Form (AQ-Short) is based on the adult self-report version (Hoekstra et al., 2011), which includes 28 items comprising two higher-order factors assessing ‘social behavior difficulties’ and ‘a fascination for numbers/patterns’. The higher order social behavior factor consists of Social Skills, Routine, Switching and Imagination. Each statement can be answered on a 4-point Likert scale (1 = definitely agree; 2 = slightly agree; 3 = slightly disagree; 4 = definitely disagree). The scoring is reversed for items in which an “agree” response is characteristic for autism (13 out of 28 items). Scores on the AQ-Short range from 28 to 112, with higher scores indicating greater endorsement of autistic traits. The AQ-Short has good sensitivity and specificity in distinguishing individuals with ASD from controls and has been shown to correlate highly with the original 50-item version of the measure (Hoekstra et al., 2011). The AQ-Short has previously been translated into Dutch (Hoekstra et al., 2008). The translation of the Chinese version of the AQ-Short was done by three experts, which followed the forward and backward procedure (Hall et al., 2017). For the parent-report version of the AQ-Short, the first-person perspective (“I”) of the items was changed into a third-person perspective (“He/She”).

Community Involvement

This study was not designed with community involvement, but the content and formulation of the NAR (Dutch) survey was developed together with a panel of individuals with ASD and their parents. The NAR also has several team members with ASD.

Statistical Analysis

Confirmatory Factor Analysis

Before conducting group comparisons, we first needed to make sure the factor structure of the parent-report AQ-Short was comparable between China and the Netherlands. The data was first cleaned, coded, and analyzed using SPSS version 22.0 for Windows (SPSS Inc., Chicago, IL). Following the factor structure identified in adult samples in Hoekstra et al. (2011), a two-factor hierarchical model was implemented in which the social skills, routine, switching, and imagination factors were predicted to load on a higher-order social behavior factor. The numbers and patterns items were predicted to load on a separate factor. This model was estimated both within the total sample and in China and the Netherlands separately. Confirmatory factor analyses (CFAs) on the parent-report AQ-Short were conducted using MPlus version 7 (Muthén & Muthén, 2012). To compare the relative fit of the models, approximation fit indices including the comparative fit index (CFI; Hu & Bentler, 1998) and Tucker-Lewis index (TLI; Tucker & Lewis, 1973) and the root mean square error of approximation (RMSEA; Steiger & Lind, 1980) were calculated. For CFI and TLI, values above 0.90 indicate a reasonable model fit, with values above 0.95 indicating a good fit to the data (Hu & Bentler, 1999; Marsh et al., 2004). For RMSEA, values below 0.06 indicate a good fit to the data (Browne and Cudeck, 1993).

Exploratory Factor Analysis

EFA analyses were conducted to further explore the underlying relationships among the items. For these analyses, WLSMV estimation with promax (i.e., oblique) rotation was applied. Norris and Lecavalier (2010) suggest that a scree plot, model fit indices, content and interpretability of the factors, salience of item loadings, items with cross-loadings, and the number of salient items in the factors should be used to ascertain the number of factors to be retained. Items with a factor loading of less than 0.32 and/or items with strong cross-loadings (loaded ≥ 0.40 on more than one factor) were dropped (Tabachnick & Fidell, 2007). Only factors with three or more strongly loading items were considered solid and justifiable factors (Costello & Osborne, 2005).

Internal Consistency and Relationships Between Factors

Cronbach’s alpha and correlations among the latent factors were estimated to measure the internal consistency and the correlations between the identified factors of the EFA model for the Chinese and the Dutch samples separately.

Group Comparison

Differences in demographics and SES characteristics across countries were evaluated using χ2 tests (for categorical variables) and t-tests (for continuous variables). The impact of demographic variables (age and sex) and SES were evaluated per country. All statistical tests were two-tailed with an alpha of 0.05. Where appropriate, these were corrected for multiple testing using the false discovery rate (FDR) procedure (Benjamini & Hohberg, 1995). Where applicable, effect sizes were calculated (Cohen, 2013). Preregistration of this study can be found at Open Science Framework (authors blinded).

Results

Table 1 presents the primary participant demographic characteristics in the Chinese and Dutch samples. The Dutch sample had a less skewed girl to boy ratio (1:3.5 compared to 1:5.5), was significantly older (t = − 16.95, p < 0.001, d = − 1.17). A smaller proportion of the Netherlands ample reported having a low-income (17.5%) compared to the Chinese sample (40.5%). In both countries, most fathers were employed, but more Dutch mothers (65.4%) were employed compared to Chinese mothers (56.9%).

Table 1 The demographic variables between China and the Netherlands
Full size table

Factor Analysis of the AQ-Short in China and the Netherlands

Confirmatory factor analyses were performed for the Chinese and Dutch samples separately. The fit indices for the Chinese AQ-Short model were: χ2 = 1378.68, df = 345, p < 0.001; RMSEA = 0.096; CFI = 0.770; TLI = 0.748. The fit indices for the Dutch CFA model were: χ2 = 1327.69, df = 345, p < 0.001; RMSEA = 0.064; CFI = 0.876; TLI = 0.864. Only the RMSEA value in the Dutch sample indicated a good model fit. The CFI, TLI values indicated moderate to poor model fit for both the Chinese and Dutch samples. Thus, there was little support for the two-factor model previously identified for the self-report adult AQ-Short. We therefore used exploratory factor analysis (EFA) to ascertain an alternate better fitting model separately for the AQ-Short in China and the Netherlands, using the same sample.

Chinese EFA Results

The scree plot for the Chinese sample suggested three or four factors. Both 3- and 4-factor models showed good fit, as indicated by RMSEA, CFI, and TLI values (see fit indices in supplementary Table 1). However, the second factor of the 4-factor model only contained one item. Taken together, the scree plot, fit values, and the number of salient items supported a 3-factor model. Item 6 “When s/he reads a story, s/he can easily imagine what the characters might look like” and item 11 “S/he finds making up stories easy” were dropped due to high cross-loadings on two factors (> 0.40). Item 17 “It does not upset him/her if his/her daily routine is disturbed” was dropped based on a low factor loading (< 0.32). Corresponding to the original AQ-Short factor model, we labeled factors 1, 2, 3 as Numbers and Patterns, Social Skills, and Mind-Reading, respectively. Final factor loadings of the 3-factor model (25 items) for the Chinese sample are presented in Table 2. Correlations among the latent factors for the Chinese sample are shown in Table 3.

Table 2 Chinese AQ-short fit indices: RMSEA = 0.051 CFI = 0.952 TLI = 0.937
Full size table
Table 3 Chinese geomin factor correlations
Full size table

Dutch EFA Results

The scree plot for the Dutch sample suggested four or five factors. Only the 4- and 5-factor models showed a good fit in the RMSEA value and the CFI and TLI values (see fit indices in Supplementary Table 2). The fourth factor “Attention Switching” and the fifth factor “Imagination” in the 5-factor model together made up the third factor in the 4-factor model. Even though in the 4-factor model, an “Attention Switching” factor and an “Imagination” factor were effectively combined into one factor, this combined factor was neither conceptually congruent with the theme of their designated factor, nor resembled any factors in the original AQ-Short models (Hoekstra et al., 2011). The five-factor Dutch model was conceptually comparable to the original AQ-Short, although the original two-factor hierarchical model previously identified in adult self-report samples was not replicated in this sample. Thus, the five-factor model was deemed the optimal model. Item 8 “In a social group, s/he can easily keep track of several different people's conversations” was eliminated based on a low factor loading (< 0.32) and item 19 “S/he enjoys doing things spontaneously” was eliminated based on high cross-loadings on two factors (> 0.40). Corresponding to the AQ-Short, we labeled factors 1, 2, 3, 4, and 5 as Imagination, Numbers and Patterns, Social Skills, Attention Switching, and Mind-reading. Final factor loadings of the 5-factor model (26 items) for the Dutch sample are presented in Table 4. Correlations among the latent factors measuring the internal relatedness of the EFA model for the Dutch sample are shown in Table 5.

Table 4 Dutch AQ-short fit indices: RMSEA = 0.043; CFI = 0.963 TLI = 0.941
Full size table
Table 5 Dutch geomin factor correlations
Full size table

Internal Consistency and Relationship Between Factors in the EFA Models

Cronbach’s alpha scores > 0.70 indicate good internal consistency (Streiner, 2003). However, when dealing with subscales derived from a single questionnaire, values around 0.60 are considered acceptable (Nunnally, 1975). In the Chinese sample, Cronbach’s α for the total AQ-Short, Numbers and Patterns, Social Skills and Mind-reading subscales were 0.77, 0.78, 0.84, and 0.65, respectively. For the Dutch sample, Cronbach’s α for total AQ-Short, Imagination, Numbers and Patterns, Social Skills, Attention Switching and Mind-reading were 0.82, 0.71, 0.76, 0.75, 0.64, and 0.62, respectively.

Overall, the relations between factors in the Chinese and Dutch EFA models ranged from very weak to moderate (0.05 to 0.43). Against expectation, a negative modest correlation (− 0.19) was found between the Social Skills and Numbers and Patterns factors in the Chinese sample, indicating that increased attention to numbers and patterns was associated with better social skills.

Impact of Age, Sex, and Socioeconomic Status (SES) Based on Country-Dependent Analysis

Due to different factor structures in the Chinese and Dutch sample, it was not possible to compare the AQ-Short total and factor scores between countries. However, we were able to examine the impact of demographic variables within both countries.

There were no significant sex differences in total AQ-Short scores within both the Chinese and Dutch samples. Dutch boys were reported to score significantly higher than girls on the Numbers and patterns factor (t = 4.60, p < 0.001, d = 0.430; males = 11.40 [SD = 3.63], females = 9.90 [SD = 3.34]). No other sex differences across the AQ-Short subscale scores were found within the Dutch sample. There were no significant sex differences found in any of the subscales within the Chinese sample.

In the Chinese sample, older children had lower total AQ-Short scores and lower Numbers and Patterns scores than young children (r = − 0.16, p = 0.04; r = − 0.14, p = 0.043). Similarly, older Dutch children had lower total AQ-Short scores (r = − 0.10, p = 0.042), Attention switching scores (r = − 0.13, p = 0.01) and Numbers and patterns scores (r = − 0.20, p < 0.001). The specific results of sex and age differences in both countries are shown in supplementary Table 3.

Within the Chinese sample, parents who were unemployed reported fewer autistic traits (AQ-Short total) and better social-skills for their child compared with employed Chinese mothers (F = 4.82, p = 0.04, η2= 0.029; F = 6.668, p = 0.01, η2 = 0.040). Chinese parents in the low income bracket also reported fewer autistic traits and better mind-reading skills in their children with ASD (F = 4.41, p = 0.04, η2= 0.028; F = 9.58, p < 0.001, η2= 0.059) than children of parents with middle and high income levels (see supplementary Table 4). In the Dutch sample, parents who were unemployed reported worse mind-reading skills for their child compared with those who were employed (F = 5.59, p = 0.03, η2 = 0.016). There was no relationship between AQ-Short subscale scores and income level in the Dutch sample (see supplementary Table 5).

Item Comparison Between China and the Netherlands

To get a deeper understanding of potential cross-cultural differences on the AQ-Short in children with ASD, the nonparametric Mann–Whitney U test (p < 0.05) was used to test for cross-cultural differences at an item level (see supplementary Table 6). Out of the 23 ‘Social behaviors’ items based on the original AQ-Short factor structure, there was a higher endorsement on nine items (39%), lower endorsement on seven items (30%) and no significant difference on the other seven items (30%) for the Chinese sample compared with Dutch participants. Out of five ‘Numbers and patterns’ items based on the original AQ-Short factor structure, the Chinese sample showed a higher endorsement of one item (20%), lower endorsement on three items (60%) and no significant difference on the other one items (20%). We also checked potential sex differences on item level across countries. Out of the 28 items, boys showed higher endorsement on four numbers and patterns items (80%, out of five items) and three social behaviors items (13%, out of 23 items), lower endorsement on five social behaviors items (22%, out of 23 items) and no significant differences on the other 16 items compared with girls. No cross-cultural sex differences on the AQ-Short item scores were found.

Discussion

This study examined and compared the factor structure of the parent-reported AQ-Short for children with ASD from China and the Netherlands. Initial CFA models showed that the AQ-Short child version did not map onto the two-factor hierarchical model previously identified in typically developing adults (Hoekstra et al., 2011) and adults with ASD (Grove et al., 2016). Subsequent exploratory factor analyses supported a three-factor model in the Chinese sample (Numbers and Patterns, Social Skills, and Mind-reading) and a five-factor model in the Dutch sample (Imagination, Numbers and Patterns, Attention Switching, Social Skills, and Mind-reading). SES, sex and age influences on AQ-Short scores were observed. Older children tended to receive lower AQ-Short scores in both countries. Chinese children from low SES families also tended to receive lower AQ scores.

Even though the EFA results showed a different factor structure for the Chinese and Dutch parent-reported AQ-Short, the Social Skills factor, Numbers and Patterns factor, and Mindreading factor emerged in both models, suggesting that the same types of behaviors tended to group together in different cultures. Consistent with the original adult AQ-Short (Hoekstra et al., 2011), a Social Skills factor and a Number and Patterns factor was found. However, the items loading on the Social Skills factor in the Chinese model included a combination of items loading on separate factors in the Dutch model (Social Skills, Attention Switching, and Imagination), indicating that certain behaviors group together differently across cultures. Specifically, Social Skills, Attention Switching and Imagination factors appear more closely linked in children with ASD in China than in the Netherlands.

How parents interpret and report on their child’s behavior is likely influenced by culture (de Leeuw et al., 2020; Norbury & Sparks, 2013). Certain AQ-Short items may therefore be interpreted differently in Eastern and Western cultures. For example, item 18 “S/he finds it easy to go back and forth between different activities”, is part of attention switching in the Netherlands, but loaded on the social skills factor in China. Possibly, in China, social skills depend more strongly on the ability to switch between activities and contexts, so as to preserve group harmony. There is also a special term, Ren Qing (人情), for social favors exchanged in the form of money, information, or affection. Item 20 “S/he finds it difficult to work out other’s thinking and feeling by looking at their face” loaded onto the Social Skills factor in the Chinese sample, but in the Dutch sample, it loaded onto the Mind-reading factor. The concept of Mind-reading, describing difficulty in perspective-taking (Baron-Cohen et al., 1985), emerged as Communication/Mindreading in Austin (2005)’s 3-factor model (Social skills, Attention to details/patterns and Mindreading/Communication). Perspective taking may depend less on the ability to decipher someone’s facial expression in China, but more so on the understanding and incorporation of the entire social context. This would be in line with earlier study findings that Asian participants attended to the whole social context when interpreting facial expressions, while American participants focused exclusively on faces (e.g., Masuda et al., 2008). Moreover, Chinese parents tended to endorse social behavior items slightly more compared to Dutch parents based on the original AQ-Short factor structure. Dutch parents more likely endorsed number and pattern items compared to Chinese parents. In sum, there may be subtle cross-cultural differences in how parents reflect on their children’s social behaviors.

Of further note is the modest negative correlation between Numbers and Patterns and Social Skills in the Chinese sample. This negative correlation has also been reported in previous studies using the AQ (Kloosterman et al., 2011; Lau et al., 2013; Ward et al., 2021). The items assessing interests in numbers and patterns may not be a good measure of autistic traits in Chinese children, as also indicated by the large proportion of Chinese parents that reported to “strongly disagree” on all of the Numbers and Patterns items (see supplementary Fig. 1). Thus, a large number of Chinese parents in this sample disagreed that their child with ASD showed a particular interest in numbers and patterns. In the Dutch sample, items 5, 7 and 13 of the Numbers and patterns factor also showed an inverse distribution. An alternative explanation of these seemingly counterintuitive results is to conceive autism as a multi-dimensional spectrum rather than a linear one. Consistent with a two-dimensional spectrum, Kitazoe et al. (2017) performed a cluster analysis of the AQ on a Japanese sample and found that participants scoring high on total AQ could be divided into two groups: one with low scores on the Attention to detail subscale but high scores on the other four subscales and the other with high scores on all five subscales. Thus, it is conceivable that some individuals with ASD may have less interest in details.

Parent-reported AQ-Short factor scores generally were the same for boys and girls in the present study, conflicting with some previous findings on sex differences in the social behavior factor of the self-report AQ-Short in adults with ASD (Hoekstra et al., 2011; Grove et al., 2016). The discrepancy between these findings and other previous adult AQ-Short studies may be (a) because women with ASD may experience more problems or are more aware of their autistic traits than men with ASD (Lai et al., 2011), or (b) parents may not recognize all autistic traits of their daughters, perhaps because girls are better at camouflaging (Jorgenson et al., 2020; Lai et al., 2016). The compensatory camouflaging may help girls receive a more similar levels of parent-reported autistic traits compared to boys. However, boys were reported to have more interest in numbers and patterns than girls in the Dutch sample, which is consistent with previous research (Grove et al., 2016; Hattier et al., 2011). The comparison between boys and girls at the item level across countries also indicated subtle biases. Boys with ASD showed a higher endorsement of the Numbers and patterns items. In both samples, parents reported fewer autistic traits and less interest in numbers and patterns for older children, suggesting that autistic symptoms and special interests may decrease with age (Waizbard-Bartov et al., 2020). However, parents of older children may also be more used to their child’s difficulties and thus underreport, despite their child potentially experiencing the same level of autistic traits as younger children. It should be noted that the reported correlations are very small. However, Bujang and Baharum (2016) found that the correlation coefficient of 0.1 can be considered important when the lowest minimum sample size is 782. The findings from these groups should be interpreted with care.

SES was associated with parent-reported autistic traits. Unemployed Chinese mothers reported fewer autistic traits and better social skills for their child than employed Chinese mothers. Unemployed Dutch mothers reported worse mindreading skills for their child than employed Dutch mothers. Furthermore, low-income Chinese families reported fewer autistic traits than middle and high-income families, which is inconsistent with previous research in Western countries showing that low SES families were more likely to endorse autistic traits (Khowaja et al., 2014; Scarpa et al., 2013). This inconsistency might be because low-income families in China are less willing to report autistic traits due to stigma and misconceptions of autism (Su et al., 2019; Sun et al., 2013). Alternatively, low-income families in China may have lower awareness and may not recognize these autistic traits in their children (de Leeuw et al., 2020). No association with family income level was found in the Dutch sample. Larger income inequality differences (using the World Bank’s Gini coefficient) and higher poverty levels in China may have resulted in more pronounced effects of SES in China compared to the Netherlands. In addition, the relatively low number of low-income families in the Dutch sample resulted in poor statistical power to detect an effect. More studies are needed to investigate the effects of SES on parent reported autistic traits. This is particularly important as autism may be underdiagnosed in low-income families in some cultural contexts (Durkin et al., 2015) and previous studies have suggested screening accuracy may be lower in low-income families (Gurthie et al., 2019).

Limitations

The current findings should be interpreted with caution. First, as participants’ intellectual ability was not formally assessed, any between-country differences may also be due to differences in cognitive ability levels. Second, the AQ-Child includes items related to children’s subjective, internal states, including imagination and mindreading, which parents may find difficult to report on. While previous studies showed adequate concurrent validity, good internal consistency and good test–retest reliability of the Mind-reading and/or Imagination subscales (Sun et al., 2019; Gomez et al., 2018), it would be worthwhile for future studies to examine differences in validity of the imagination/mindreading subscales and the social behavior subscales. Furthermore, study recruitment was not the same in both countries. Chinese participants were recruited via various services for children with ASD around China and most of these participants were identified at a younger age. In contrast, Dutch data were collected through an online volunteer register, including children with ASD who were not necessarily under treatment or assessment in clinics. Behavior of children receiving services may be rated differently than those who are not under treatment in clinics. Moreover, as only children with autism were examined, it is uncertain whether the factor structure differences are specific to Chinese children with autism or reflect more general cross-cultural differences. For a better understanding of cross-cultural differences, it would be worthwhile to examine which items of the AQ-Short best discriminate between clinical and non-clinical groups across cultures. As such, this study’s outcome should be taken into consideration as preliminary.

Conclusion

In summary, the factor structure of the parent-reported AQ-Short was substantially different between China and the Netherlands, suggesting that interpretation and reporting of autism symptoms may be culture-dependent. Differences in item-factor loadings indicate variation in the reporting of autistic traits between parents from China and the Netherlands, and this may hamper cross-cultural comparisons. This indicates that the Chinese version of the AQ-Short may assess slightly different autistic traits than the Western version. It would be worthwhile for future research to interview Chinese parents of children with ASD from low-income families to better explore how they recognize, interpret and report their child’s autistic traits. This study is the first to test the factor structure of the parent-report AQ-Short across cultures. While our findings require replication in wider samples, including children without ASD, we were able to show that the items of the AQ-Short generally detect autistic traits equivalently in boys and girls with ASD. Autistic traits and special interests may decrease with age, and this decrease may be similar across countries. However, longitudinal studies are needed to further examine the trajectories of autism severity change during childhood. Furthermore, low SES Chinese families reported fewer autistic traits, making them vulnerable for under-detection, which was not the case in our Dutch participants. The under-reporting of autistic traits in low SES families deserves further research to address culture-specific disparities in autism diagnoses and service access.